TWI703997B - System for management of diabetes - Google Patents

Abstract

Described and illustrated is a system for management of diabetes that includes an infusion pump, glucose sensor and controller with a method programmed in the controller. The infusion pump is configured to deliver insulin. The glucose sensor senses glucose levels in the subject and provide output signals representative of the glucose levels in the subject. The controller is programmed receives signals from at least one of the glucose sensor and the pump and configured to issue signals to the pump to deliver an amount of insulin determined by a feedback controller that utilizes a model predictive control based on desired glucose levels, insulin amount delivered and measured glucose levels of the subject. The controller is also configured to deliver insulin using a tuning factor (R) for a model predictive controller in the microcontroller as a conservative setting otherwise the system maintains a current tuning factor (R) for the controller.

Description

Diabetes Management System

Diabetes is a metabolic-related chronic disease caused by the pancreas's inability to produce sufficient hormone insulin, which reduces the body's ability to metabolize glucose. This insufficient pancreas function leads to hyperglycemia, that is, there is a very large amount of glucose in the plasma. Intractable hyperglycemia and/or hypoinsulinemia have been associated with various severe symptoms and fatal long-term complications, such as dehydration, ketoacidemia, diabetic coma, cardiovascular disease, chronic renal failure, retinal damage, and nerves Damage, with the risk of amputation. Because the restoration of endogenous insulin production is not yet feasible, permanent treatment is needed to provide constant glucose control so that blood glucose levels are always maintained within normal limits. This type of blood glucose control is achieved by regularly supplying insulin from the outside to the patient's body to thereby reduce the elevated blood glucose level.

Exotic biological products such as insulin are usually administered by multiple injections per day, which are injected with a mixture of fast-acting and intermediate-acting drugs via a hypodermic syringe. It was found that because the delivery is different from the production of physiological hormones (physiological hormones enter the bloodstream at a slower rate and longer period), this injection method cannot achieve the best degree of blood sugar control. Improved blood sugar control can be achieved by so-called intensive hormone therapy. Intensive hormone therapy is based on multiple daily insulin injections, including one or two injections a day to provide long-acting hormones that provide benchmark hormones and additional injections before each meal. Fast-acting hormones in proportion to the amount of meals. Although traditional syringes have been at least partially replaced by insulin pen needles, frequent injections are still extremely inconvenient for patients, especially for patients who cannot reliably manage their injections.

Through the development of drug delivery devices, the need for syringes and drug pens and multiple injections per day have been reduced to substantially improve the treatment of diabetes. Drug delivery The device allows the delivery of drugs in a manner most similar to the natural occurrence of physiological processes, and can control the drug delivery device to comply with standards or personally modified agreements, giving patients better blood sugar control.

In addition, the drug delivery device can be directly delivered to the intra-abdominal space or delivered through the vein. The drug delivery device can be constructed as an implantable device for placement under the skin, or can be constructed as an external device containing an infusion sleeve for delivery of transdermal drugs (such as through a patch) via a percutaneously inserted catheter, cannula, or subcutaneously Inject into the patient. The external drug delivery device is hung on clothes, hidden under or inside the clothes, or hung on the body, and is usually controlled via a user interface built into the device or on a separate remote control device.

Achieving acceptable blood glucose control requires blood or inter-tissue glucose monitoring. For example, the delivery of an appropriate amount of insulin through a drug delivery device requires the patient to frequently determine their own blood glucose level and manually enter this value into the user interface of an external pump, and then the pump calculates the default or currently in-use insulin delivery agreement (ie , Dosage and timing), and then communicate with the drug delivery device to adjust its operation accordingly. Generally, an intermittent measuring device (such as a handheld electronic meter) receives a blood sample through an enzyme-based test strip and calculates a blood glucose level based on the enzyme reaction to determine the blood glucose concentration.

In the past 20 years, continuous glucose monitoring (CGM) has been used in conjunction with drug delivery devices to allow insulin injection into closed-loop control of diabetic patients. In order to allow closed-loop control of insulin injection, a proportional integral derivative ("PID") controller has been used in conjunction with a mathematical model of the metabolic interaction between glucose and insulin in humans. The PID controller can be tuned based on the simple rules of the metabolic model. However, when the PID controller is tuned or configured to actively adjust the blood glucose level of the subject, a set level overshooting may occur, and fluctuations usually follow, which is extremely undesirable in regulating blood glucose. Alternative controllers have been studied. It has been determined that the Model Predictive Controller ("MPC") used in the petrochemical industry (where the processing requires a large time delay and system response) is most suitable for the complex interaction between insulin, glucagon and blood sugar. The effect of MPC controller is more robust than PID, because MPC considers the short-term effects of control changes and the constraints when determining MPC output, while PID usually only involves the past output when determining future changes. Restrictions can be implemented in the MPC controller so that the MPC prevents the system from losing control when the limit has been reached. The other of MPC controller One advantage is that, in some cases, the model in MPC theoretically compensates for dynamic system changes, while feedback control such as PID control cannot have such dynamic compensation.

Therefore, MPC can be regarded as a combination of feedback and feedforward control. However, MPC usually requires a metabolic model to mimic the interaction between insulin and glucose in biological systems as much as possible. Therefore, due to the biological differences between people, the current MPC model continues to be further improved and researched. The MPC information background about the MPC controller, the differences in the MPC, and the details of the mathematical model representing the complex interaction between glucose and insulin, are all shown and described in the following documents: US Patent No. 7,060,059; US Patent Application No. 2011/0313680 and 2011/0257627; International Publication WO 2012/051344; Percival et al. " Closed-Loop Control and Advisory Mode Evaluation of an Artificial Pancreatic β Cell: Use of Proportional-Integral-Derivative Equivalent Model-Based Controllers "("Journal of Diabetes Science and Technology, Vol.2, Issue 4, July 2008"); Paola Soru et al. " MPC Based Artificial Pancreas; Strategies for Individualization and Meal Compensation "("Annual Reviews in Control 36, p .118-128(2012)"); Cobelli et al. " Artificial Pancreas: Past, Present, Future "("Diabetes Vol.60, November 2011"); Magni et al. " Run-to-Run Tuning of Model Predictive Control for Type 1 Diabetes Subjects: In Silico Trial "("Journal of Diabetes Science and Technology, Vol.3, Issue 5, September 2009"); Lee et al. " A Closed-Loop Artificial Pancreas Using Model Predictive Control and a Sliding Meal Size Estimator "("Journal of Diabetes Science and Technology, Vol.3, Issue 5, September 2009"); Lee et al. " A Closed-Loop Artificial Pancreas bas ed on MPC: Human Friendly Identification and Automatic Meal Disturbance Rejection ^{"(" Proceedings of the 17 th World} Congress, The International Federation of Automatic Control, Seoul Korea, 2008 July 6 to 11 "); Magni et al.," Model Predictive Control of Type 1 Diabetes: An in Silico Trial "(" Journal of Diabetes Science and Technology , Vol.1, Issue 6, November 2007 "); Wang et al.," Automatic Bolus and Adaptive Basal Algorithm for the Artificial Pancreatic β - Cell "("Diabetes Technology and Therapeutics, Vol.12, No.11, 2010"); and Percival et al. " Closed-Loop Control of an Artificial Pancreatic β -Cell Using Multi-Parametric Model Predictive Control "("Diabetes Research 2008 ").

All articles or documents listed in this application are hereby incorporated into this application by reference, as if they were fully described herein.

The applicant invented a technology that allows tuned model predictive control, which enables the model predictive controller implemented in the microcontroller of our system to qualitatively compensate for the inherent inaccuracy of the continuous glucose sensor. In particular, a method for controlling an infusion pump using a microcontroller to control the pump and receive data from at least one glucose sensor is provided. The method can be implemented by: at each time interval of a series of discrete time interval indexes ("k"), the glucose level of the subject is measured from the glucose sensor to provide at least one glucose measurement value; Within the time interval of, determine whether the glucose sensor is replaced by a new glucose sensor; if the determination step is true, set a tuning factor (R) for the model predictive controller in the microcontroller It is a conservative setting; otherwise, if the determination step is false, the current tuning factor (R) of the controller is maintained; based on the model predictive controller using the tuning factor (R), it is calculated by the microcontroller The amount of insulin used for delivery, so as to provide a calculated amount of insulin to be delivered to the subject in one or more discrete time intervals; and delivery is determined by the calculation step by starting a pump by the microcontroller The amount of insulin.

In yet another aspect, a system for managing diabetes is provided, which includes an intermittent glucose meter, a continuous glucose meter, and an infusion pump coupled to a controller. The intermittent glucose meter is configured to measure the subject's blood glucose and raise blood glucose at discrete and non-uniform time intervals. Provide this intermittent blood glucose level as a calibration. The continuous glucose monitor is configured to continuously measure the glucose level of the subject at discrete and generally consistent time intervals, and to provide the glucose level in the form of glucose measurement data at each interval. The insulin infusion pump is used to deliver insulin. The microcontroller is in communication with the pump, the glucose meter, and the glucose monitor. Specifically, after replacing the continuous glucose monitor with a new continuous glucose monitor, the controller sets a tuning factor (R) to a conservative value for a predetermined time period, so that the controller is based on the conservative tuning factor The model predictive control determines the insulin delivery rate for each time interval in the time interval index (k), and commands the pump to deliver at a predetermined insulin delivery rate.

In each of the above aspects, the following features can also be used in combination with each aspect. For example, at least one glucose sensor may include both a continuous glucose sensor and an intermittent glucose meter; the conservative tuning factor (R) may include a value of about 500, and the predetermined time period may include a glucose sensor About 24 hours after the sensor is replaced by a new glucose sensor; the aggressive tuning factor (R) may include a value of about 10; the aggressive tuning factor (R) may include a value of about 1000; the model The predictive controller has an error weighting factor (Q) related to the tuning factor (R), where:

Wherein R may include the tuning factor; Q may include an error weighting factor.

Those skilled in the art should clearly understand these and other embodiments, features and advantages when referring to the following more detailed descriptions of various exemplary embodiments of the present invention together with the accompanying drawings briefly described first.

10: Control logic module; MPC logic module

12: desired glucose concentration; glucose concentration range

14: The first output; the control signal of the insulin pump

15: Predicted glucose value

16: insulin pump

18: The amount of insulin required

20: Object

22: Continuous glucose sensor

26: Difference

28: Update filter

29: Intermittent glucose meter

30: Calibration index module

32: data omission module

34: tuning factor

36: flag or switch

100: Drug Delivery System

102: Drug delivery device; insulin delivery device

104: remote controller; microcontroller; remote microcontroller

106: Infusion group

108: Flexible tube

112: RF communication; CGM sensor

114: Intermittent blood glucose meter

116: Remote health monitoring station; RF module

118: wireless communication network

126: Personal computer or network computer

128: server

200: schematic diagram

302: Step

304: Step

306: Step

308: step

310: Step

312: Step

314: Step

316: Step

A: Control junction

B: Control junction

The drawings incorporated herein and forming part of this specification illustrate the presently preferred embodiments of the present invention, and together with the prior art given above and the embodiments given below are used to explain the features of the present invention (wherein The same numbers indicate the same elements).

Figure 1 shows the system, where the controller for the pump or glucose monitor is separate from the infusion pump and the glucose monitor, and where a network can be coupled to the controller to provide near real-time monitoring.

Fig. 2A shows in schematic form an exemplary embodiment of a diabetes management system.

Figure 2B shows a graph of glucose values at a 48-hour time interval (or K equals 12 in each hour of the CGM signal), where other events (for example, missing CGM data, sensor replacement or calibration measurement) are superimposed On the graph of the glucose value.

Figure 2C shows the tuning factor at a time interval of 48 hours (or the time interval or exponent where K is equal to 12 in each hour of the CGM signal), where the tuning factor R will be due to missing CGM data, sensor replacement and calibration measurement And change.

Figure 3 shows the logic used in the controller in Figure 1 or Figure 2A.

The following detailed description must be read with reference to the drawings, where similar elements in different drawings are numbered in the same way. The drawings are not necessarily drawn to scale, they depict selected embodiments and are not intended to limit the scope of the invention. This detailed description explains the theory of the present invention by way of example rather than limitation. This description enables those skilled in the art to make and use the present invention, and it describes several embodiments, changes, variations, substitutions, and uses of the present invention, including those currently believed to be the best mode for carrying out the present invention.

As used in this specification, the terms "about" or "abbreviated" used in any value or range refer to appropriate scale tolerances so that a part or assembly of components can be targeted for its intended purpose (as in this specification) Narrator) play a role. In addition, as described in this article, the terms "patient", "host", "user" and "object" refer to any human or animal subject, and it is not intended to limit these systems and methods to human use, even if The application of the present invention to human patients represents a preferred embodiment. In addition, the term "user" includes not only patients who use the drug infusion pump device, but also caregivers (such as parents or guardians, caregivers or family care workers). The term "medicine" may include hormones, biologically active materials, medicines, or chemicals that cause a biological reaction (for example, a glycemic reaction) in the body of the user or patient.

FIG. 1 shows a drug delivery system 100 according to an exemplary embodiment using the principles of the present invention. The drug delivery system 100 includes a drug delivery device 102 and a remote controller 104. The drug delivery device 102 is connected to the infusion set 106 via a flexible tube 108.

The drug delivery device 102 is configured to transmit data to and receive data from the remote controller 104 via radio frequency communication 112, for example. The drug delivery device 102 can also operate as a standalone device and has its own built-in microcontroller. In one embodiment, the drug delivery device 102 is an insulin injection device, and the remote controller 104 is a handheld portable controller. In this embodiment, the data transmitted from the drug delivery device 102 to the remote controller 104 may include, for example, insulin delivery data, blood glucose information, baseline insulin, single dose insulin to carbohydrate ratio, or insulin sensitivity factor, etc. And other information. The microcontroller 104 is configured to include the MPC controller 10, which is programmed to receive continuous glucose readings from the CGM sensor 112. The data transmitted from the remote microcontroller 104 to the insulin delivery device 102 may include glucose test results and a food database, so that the drug delivery device 102 can calculate the amount of insulin to be delivered by the drug delivery device 102. Alternatively, the remote microcontroller 104 can perform basic administration or single dose calculations, and send the results of such calculations to the drug delivery device. In an alternative embodiment, the intermittent blood glucose meter 114 may be used alone or in combination with the CGM sensor 112 to provide data to either or both of the microcontroller 104 and the drug delivery device 102. Alternatively, the remote microcontroller 104 and the meter 114 can be combined into (a) an integrated single device; or (b) two separate devices that can be connected to each other to form an integrated device. Each of the devices 102, 104, and 114 has a suitable microcontroller (not shown in the figure for the sake of brevity) programmed to perform various functions.

The drug delivery device 102 can also be configured to perform two-way wireless communication with the remote health monitoring station 116 via, for example, the wireless communication network 118. The remote controller 104 and the remote monitoring station 116 can be configured to perform two-way wired communication via, for example, a telephone fixed communication network. For example, the remote monitoring station 116 can be used to download the upgrade software to the drug delivery device 102 and process the information from the drug delivery device 102. Examples of remote monitoring stations 116 may include (but are not limited to) personal computers or network computers 126, servers 128 to memory storage, personal digital assistants, other mobile phones, hospital base monitoring stations or dedicated Remote clinic monitoring station.

The drug delivery device 102 includes electronic signal processing components, including a central processing unit and a memory element for storing control programs and operating data, for transmitting communication signals (ie, messages) to and receiving from the remote controller 104 The radio frequency module 116 for communication signals (ie messages), a display for providing operation information to the user, a plurality of browsing buttons for the user to input information, a battery for providing power to the system, and a battery for providing An alarm (for example, visual, auditory, or tactile) feedback to the user, a vibrator for providing feedback to the user, and for forcing insulin from an insulin reservoir (for example, an insulin cartridge) to pass through the infusion set 108/ 106 side port and enter the drug delivery mechanism (for example, drug pump and driving mechanism) in the user's body. An example of the drug delivery device 102 (or pump 16) may be in the form of an improved Animas Vibe insulin pump manufactured by Animas Corporation (Wayne, Pennsylvania, USA).

The CGM sensor 112 can be used to determine the glucose level or concentration. The CGM sensor 112 uses amperometric electrochemical sensor technology to measure glucose with three electrodes operatively connected to the sensor electronics and is covered by a sensing film and a biological interface film attached by a clip.

The top of the electrode contacts the electrolyte phase (not shown in the figure), which is a free-flowing fluid phase arranged between the sensing membrane and the electrode. The sensing film may include an enzyme covering the electrolyte phase, for example, glucose oxidase. In this exemplary sensor, counter electrodes are provided to balance the current generated by the species measured at the working electrode. For glucose sensors based on glucose oxidase, the species measured at the working electrode is H ₂ O ₂ . The current generated at the working electrode (which flows through the circuit to the counter electrode) is proportional to the diffusion flux of H ₂ O ₂ produced by this electrochemical conversion of glucose into its own enzyme by-product. According to this, an original signal representing the glucose concentration of the user's body can be generated, so the original signal can be used to estimate a meaningful glucose value. The detailed information of the sensor and associated components is shown and described in US Patent No. 7,276,029, which is incorporated into this application by reference, as if fully described herein. In one embodiment, the exemplary embodiment described herein may also utilize a continuous glucose sensor from Dexcom Seven System® (manufactured by Dexcom Inc.).

In an embodiment of the present invention, the following components can also be used as a diabetes management system similar to an artificial pancreas: OneTouch Ping® of Animas Corporation A glucose management system, which includes at least one infusion pump and an intermittent glucose sensor; and DexCom® G4 Platinum® CGM from DexCom Corporation, which has an interface for connecting these components and programming in MATLAB® language and connecting these components to Together with the accessory hardware; and the control algorithm in the form of MPC, which automatically adjusts the insulin delivery rate based on the patient's glucose level, historical glucose measurement values and predicted future glucose trends as well as patient-specific information.

2A shows a schematic diagram 200 of the system 100 of FIG. 1 programmed with a solution designed by the applicant to counter the undesired effects of the closed loop control system. In particular, FIG. 2A provides an MPC, which is programmed into the control logic module 10 used in the controller 104. The MPC logic module 10 receives the desired glucose concentration or glucose concentration range 12 (along with any modifications from the update filter 28 in order to be able to maintain the object's output (ie, the glucose level) within the desired range).

2A, the first output 14 of the MPC-enabled control logic 10 can be the control signal of the insulin pump 16, which delivers the required amount of insulin 18 to the subject 20 at predetermined time intervals, which can use the time interval index k every 5 minutes Incorporate into the index. The second output in the form of a predicted glucose value of 15 can be used to control junction B. The continuous glucose sensor 22 (or 112 in FIG. 1) measures the glucose level of the subject 20 to provide a signal 24 representing the actual or measured glucose level to the control junction B, and the control junction B is between all The difference between the measured glucose concentration 24 and the measured glucose concentration predicted by MPC. This difference provides input 26 for the update filter of the model state variables. The difference 26 is provided to an estimator (also called an update filter 28), which provides an estimate of model state variables that cannot be directly measured. The update filter 28 is preferably a progressive filter in the form of a Kalman filter containing model tuning parameters. The output of the update or pass filter 28 is provided to the control junction A, and the MPC in the control logic 10 uses the output of the control junction A to further improve the control signal 14 to the pump 16 (or 102 in FIG. 1). The MPC controller 10 uses a tuning factor 34 to "tune" the insulin delivery of the controller. To achieve this, the applicant has designed a calibration index module 30 and a data omission module 32 to adjust the tuning factor. The calibration index module 30 is configured to track the calibration data of the glucose measurement value, which is typically completed by an intermittent glucose monitor, such as a blood glucose test strip and a measurement system. The data omission index module 32 is configured to track the number of measurement values or data missing from the continuous glucose sensor 22.

A brief overview of the MPC used for the control logic 10 as described above is set forth here. The MPC logic system is formulated to control the target's glucose level to a safe glucose zone. The lower limit of blood glucose in the zone changes between 80-100mg/dL and the upper limit of blood glucose changes between about 140-180mg/dL; this algorithm is called " Zone MPC". Control to the target zone is generally used for control systems lacking a specific set point, and the purpose of the controller is to maintain a controlled variable (CV) such as glucose value in a predefined zone. Control to zone (ie, euglycemic zone) is very suitable for artificial pancreas, because artificial pancreas has no natural blood glucose set point. In addition, the inherent advantage of controlling to the zone is to limit the pump's ability to actuate/activate, so that if the glucose level is within the zone, no additional correction is required.

In the instant condition, the rate of insulin delivery zone MPC Law Department of I _D is calculated from the optimized line rate of insulin delivery that evaluates the next at each sampling time. The optimization at each sampling time is based on the estimated metabolic state (plasma glucose, subcutaneous insulin) obtained from the dynamic model stored in the module 10.

The MPC of Control Logic 10 incorporates a clear model of human T1DM glucose-insulin dynamics. The model is used to predict the future glucose value and calculate the future controller movement that will make the glucose change within the desired range. The MPC of the controller can be formulated for both the discrete time system and the continuous time system; the controller is set in discrete time, where the discrete time (stage) index k refers to the k ^th sampling period that occurs in continuous time, where minutes It is the sampling period. Software constraints ensure that islets t=k. The T _s element delivery rate is limited by T _s =5 between the minimum value (ie, zero) and the maximum value. Then implement (out of N steps)

The first insulin injection. In the next step k +1, the procedure is repeated based on the newly measured glucose value and the last insulin rate.

Specifically, we start with the original linear difference model for zone MPC: G '( k ) = a ₁ G '( k -1)+ a ₂ G '( k -2)+ a ₃ G '( k -3)+ a ₄ G '( k -4)+ a ₅ G '( k -5)+ bI _M ( k -4) I _M ( k ) = c ₁ I _M ( k -1)+ c ₂ I _M ( k -2)+ d ₁ I ' _D ( k -1)+ d ₂ I ' _D ( k -2) Equation (1)

Among them: k is a discrete time interval index, which has a series of index counters, where k=1, 2, 3...

G’ is the measured glucose concentration

I _M is "mapping insulin", which is not a measured amount

I _'D to deliver insulin or manipulated variables

And coefficients a ₁ ~2.993; a ₂ ~(-3.775); a ₃ ~2.568; a ₄ ~(-0.886); a ₅ ~0.09776; b~(-1.5); c ₁ ~1.665; c ₂ ~(- 0.693); d ₁ ~0.01476; d ₂ ~0.01306.

Using the metabolic simulator known to the FDA accepted by the skilled person, equation (1) can be simplified to the following equation (2) linear difference model: ( a ) G '( k )=2.993 G '( k -1)-3.775 G '( k -2)+2.568 G '( k -3)-0.886 G '( k -4)+0.09776 G '( k -5)-1.5 I _M ( k -4)+0.1401 Meal _M ( k -2) +1.933 Meal _M ( k -3)

( b ) I _M ( k )=1.665 I _M ( k -1)-0.693 I _M ( k -2)+0.01476 I _D '( k -1)+0.01306 I _D '( k -2)

( c ) Meal _M ( k ) = 1.501 Meal _M ( k -1) + 0.5427 Meal _M ( k -2) + 0.02279 Meal ( k -1) + 0.01859 Meal ( k -2) (2)

Wherein: G 'is the output of the glucose concentration (G) the deviation variables (mg / dL), i.e. G' ≡ G -110mg / dL, I D ' input (I _D) deviation variables (U / h) insulin injection rate, That is, I _D' ≡ I _D -benchmark U/h, meal is CHO intake input (g-CHO), I _M is the mapped subcutaneous insulin injection rate (U/h), and meal _M is mapped CHO Intake input (gram-CHO).

Dynamic model with insulin injection rate in the formula (2) (I _D) and plasma glucose uptake in CHO input (meal) related effects. This model represents a single average model of the total ethnic group of objects. The model and its parameters are fixed.

The second-order input transfer function described by part (b) and (c) of equation (2) is used to generate artificial input memory in the zone MPC architecture to prevent insulin overdose and thus prevent hypoglycemia. In order to prevent over-delivery of insulin, any subsequent estimation of insulin delivery must take into account the effects of the previously administered insulin and insulin. However, the relatively low-level single-state linear difference model uses the output (blood sugar) as the main source of "memory" for the input (insulin) previously administered. This may result in insufficient or excessive insulin delivery due to model mismatches, noise, or changes in the subject’s sensitivity to insulin. This mapping is used by increasing insulin and meals carry two additional states entered a long memory of insulin (I _M and meals _M) and slow down.

，M.D.、及Francis J.Doyle III，Ph.D.之「Zone Model Predictive Control：A Strategy to Minimize Hyper and Hypoglycemic Events」，Journal of Diabetes Science and Technology，卷4，期號4，2010年七月；及2011年8月25日公開之美國專利申請案公開號第2011/0208156號，作者為Doyle等人，標題為「Systems,Devices,and Methods to Deliver Biological Factors or Drugs to a Subject」，所有上述均併入本文中作為參考，如同本文所闡述，並提供附本於附錄中。區帶MPC的額外細節係顯示及描述於美國專利申請案公開號第20110208156號中，其特此以引用方式併入，並提供附本於附錄中。Maciejowski JM於「PREDICTIVE CONTROL WITH CONSTRAINTS」Harlow，UK：Prentice-Hall，Pearson Education Limited，2002)中提出相關之區帶MPC衍生。區帶MPC係藉由定義固定上界限與下界限作為軟約束實施，即當該預測之受控制變數(CV)係在所要區帶之中或之外時分別使最適化權值在零與某些最終值之間切換。預測的殘值通常定義為該超出所要區帶之受控制變數(CV)與最近界限之間的差值。區帶MPC通常被劃分成三個不同區帶。准許之範圍為控制目標並且係由上界限與下界限定義。上區帶表示非所要之高預測血糖值。下區帶表示非所要之低預測血糖值，其表示為低警示區帶之低血糖區帶或前低血糖保護區。區帶MPC 藉由依據指定之約束限制來操控近期胰島素控制移動以保持在准許之區帶中來使預測糖血症最適化。 The zone MPC is applied when the specific set point value of the controlled variable (CV) is low relative to the zone defined by the upper and lower limits. In addition, it is not practical to use a fixed set point when there is noise and model mismatches. Develop regional MPC through research at the University of California at Santa Barbara and Sansum Diabetes Research Institute. Other details derived from zone MPC technology are shown and described in Benyamin Grosman, Ph.D., Eyal Dassau, Ph.D., Howard C. Zisser, MD, Lois Jovanovi

" Zone Model Predictive Control: A Strategy to Minimize Hyper and Hypoglycemic Events ", MD, and Francis J. Doyle III, Ph.D., Journal of Diabetes Science and Technology, Volume 4, Issue 4, July 2010; And US Patent Application Publication No. 2011/0208156 published on August 25, 2011, authored by Doyle et al., entitled " Systems, Devices, and Methods to Deliver Biological Factors or Drugs to a Subject ", all of the above Incorporated into this article as a reference, as described in this article, and attached to the appendix. Additional details of zone MPC are shown and described in U.S. Patent Application Publication No. 20110208156, which is hereby incorporated by reference, and an attachment is provided in the appendix. Maciejowski JM proposed related zonal MPC derivation in " PREDICTIVE CONTROL WITH CONSTRAINTS" Harlow, UK: Prentice-Hall, Pearson Education Limited, 2002). Zone MPC is implemented by defining fixed upper and lower bounds as soft constraints, that is, when the controlled variable (CV) of the prediction is in or outside the desired zone, the optimal weights are respectively zero and a certain value. Switch between these final values. The predicted residual value is usually defined as the difference between the controlled variable (CV) outside the desired zone and the nearest limit. Zone MPC is usually divided into three different zones. The permitted range is the control target and is defined by the upper limit and the lower limit. The upper band represents the undesired high predicted blood glucose level. The lower zone represents the undesired low predicted blood glucose value, which is represented as the hypoglycemia zone of the low warning zone or the former hypoglycemia protection zone. Zone MPC optimizes the prediction of glycemia by manipulating recent insulin control movements according to specified constraints to stay in the permitted zone.

The core of zone MPC lies in the value function formula that covers the zone formula. Like other forms of MPC, zone MPC predicts future output by using a clear model of past input/output records and future input movements that need to be optimized. However, optimization attempts to maintain or move the predicted output to the zone defined by the upper and lower limits, rather than driving to a specific fixed set point. The linear difference model is used to predict the blood glucose dynamics, and the optimization is based on the weight defined in the constraint and its value function to reduce the future blood glucose deviation zone.

The zone MPC value function J used for the proposed work is defined as follows:

Or for our application: J ( I _D ' )=Σ∥ G ^zone ( k + j )∥+ R． _{Σ∥ I D (k + j)} - reference (k + j) ∥ (4 )

Wherein J is a cost function; Q is the prediction weighting factor glucose entry; R & lt is the cost function in future recommendations input tuning factor; F prediction function (in the formula (2)); and a vector I _D comprising the set of Recommended recent insulin injection. It is a "manipulated variable" because it is manipulated to find the minimum value of J.

其中血糖區帶係由上限G _ZH 及下限G _ZL 定義。 The G ^zone is a variable that quantifies the deviation of the future model-predicted CGM value G outside the specific blood glucose zone and is determined by the following comparison:

The blood glucose zone is defined by the upper limit G _ZH and the lower limit G _ZL .

Therefore, if all the predicted glucose values are in the zone, each component of the G ^zone is equal to 0, so J is minimized, and for the time of the day I _D = the benchmark , that is, the algorithm "defaults" to disease Suffer from the current baseline insulin injection rate. On the other hand, if any predicted glucose value is outside the zone, then the G ^{zone is} > 0, and therefore "contributes" to the value function. In this case, the recent recommendations of insulin injection quantity I _D is derived from a reference in order to prevent the occurrence of the G ^zone once ^{with a} deviation beyond the zone of z, it will also "contribute" to the value of the function. Then a quantitative balance is found in the optimization based on the weighting factor R.

R ^M ，例如，若M=5，各最適化的初始猜測為I _D ’=[0 0 0 0 0]。此意味著初始猜測等於基準速率。 In order to solve the optimization problems of equations (2) to (5), commercially available software (for example, MATLAB's "fmincon.m" function) is used. For this function, the following parameters optimized for each: ○ zero vectors for insulin delivery rate of I _D '(0) of the initial guess

R ^M , for example, if M = 5, the initial guess of each optimization is I _D ' =[0 0 0 0 0]. This means that the initial guess is equal to the base rate.

○ The maximum number allowed for function calculation is Max_f =100* M , where M is the control level previously described.

○ The maximum number of iterations is Max_i = 400, which is fixed.

○ Term_cost =1e-6, the termination of the value function value, which is fixed.

○ manipulating variable I _D 'of the termination tolerance Term_tol 1e-6.

其中基準為對象的基準速率，如由該對象或其醫師所設定，預期位於約0.6至1.8U/hr的範圍內。 The following constraints based strictly embodiment the actuating variable (I _D '):

The benchmark is the benchmark rate of the subject, which is expected to be in the range of about 0.6 to 1.8 U/hr, as set by the subject or its physician.

在較佳的實施例中，吾人使用M=5及P=108的標稱值。 Although the value of the control level parameter M and the value of the predicted level parameter P have a significant effect on controller performance and are generally used to tune MPC-based controllers, they can be tentatively tuned based on system knowledge. The tuning rules are known to those skilled in the field. According to these rules, M and P can vary between the following ranges:

In a preferred embodiment, we use nominal values of M =5 and P =108.

在較佳的實施例中，吾人可使用R/Q=500的標稱值。 The ratio of the output error weighting factor Q and the input change weighting matrix or tuning factor R can be changed between the following ranges:

In a preferred embodiment, we can use a nominal value of R/Q =500.

Once the controller 10 is initialized and turned on in the microcontroller or 104 of the device 102, real-time calculations are performed every five minutes, which is consistent with the sampling time for the continuous glucose sensor 22. Delivering a first component I _D as the dose of insulin to the patient through an insulin pump 16, five minutes later, a new CGM readings obtained, and the procedure is repeated. Note that the future control movement is restricted by strict constraints, which are set by the ability of the insulin pump to deliver the maximum rate of insulin and the ability to deliver negative insulin values. Other detailed information on topics related to state estimators and other MPCs are provided by Rachel Gillis et al. " Glucose Estimation and Prediction through Meal Responses Using Ambulatory Subject Data for Advisory Mode Model Predictive Control "("Journal of Diabetes Science and Technology Vol. 1, Issue 6, November 2007) and Youqing Wang et al. " Closed-Loop Control of Artificial Pancreatic β- Cell in Type 1 Diabetes Mellitus Using Model Predictive Iterative Learning Control "("IEEE Transactions on Biomedical Engineering, Vol. 57, No .2, February 2010"), these documents are hereby incorporated into this application by reference, as if fully described here.

Known tuning parameters or tuning factors (designated "R" here) can have a significant impact on the quality of glucose control. Parameters (called aggressive factors, gains, and other names) determine how quickly the algorithm responds to changes in glucose concentration. The relatively conservative value of R causes the controller to be slower than adjusting the insulin injection volume in response to changes in glucose (relative to the benchmark); on the other hand, the relatively aggressive value of R causes the controller to respond quickly to changes in glucose. In principle, an aggressive controller will result in the best glucose control if 1) the available glucose measurements are accurate and 2) the model prediction of future glucose trends is accurate. If these conditions are not established, it is safer to use a conservative controller.

The tuning factor R (referred to here as 34) is worthy of discussion at this point in Figure 2A. Because it is not straightforward to determine the accuracy of the glucose measurement value on the line or the accuracy of the model prediction, it is difficult to know the appropriate tuning factor 34 or R for the current conditions. However, under certain circumstances, when the continuous glucose monitor (CGM) has become "less accurate" or "more accurate" in a qualitative sense, it may be determined with a high degree of certainty.

The applicant believes that it is reasonable to assume that the CGM may become less accurate when each sample is calibrated and suddenly become less accurate when it cannot report glucose measurements. Missing or omitted CGM measurements can be due to the uncertainty of the glucose trend (a feature of the CGM algorithm) or simply due to radio frequency communication problems. Similarly, it is reasonable to assume that the CGM has become more accurate after calibration and the reported glucose value will become more accurate after one or more missed readings are re-established.

The applicant therefore found that it is possible to use these insights to automatically tune the controller, that is, when the CGM becomes less accurate, use a more conservative R value for the tuning factor, and when the CGM becomes more accurate, use a more aggressive value. This is explained in the general sense as follows.

Make the radical value or tuning factor R limited by the constants Rmin (for the most radical controller) and Rmax (for the most conservative controller): Rmin<=R<=Rmax Equation (9)

There are two nominal increments (strictly positive): r ₁ is a relatively small increment, which is related to the accuracy of the CGM about how recent the calibration is, and r ₂ is a larger increment , Which is related to the accuracy of the CGM about the missing sample.

The tuning factor to be used in the MPC calculation at the current sampling instant is R(k) (indexed by k), where R(k) is automatically updated in each sample based on a nominal R value (R _NOM ) Generally speaking, the nominal R value for each user can be different: R(k)=R _NOM +r ₁ CAL(k)+r ₂ *MISSED(k), formula (10)

And R _MIN <=R(k)<=R _MAX formula (11)

CAL(k)=kk _CAL -6, for kk _CAL >=6 formula (12)

CAL(k)=kk _CAL2 -6, for kk _CAL <6 formula (13)

R_MAX

100* R_NOM Where: R _NOM

R _MAX

100* R _NOM

R_MIN

R_NOM R _NOM /100

R _MIN

R _NOM

r ₁

(R_MAX-R_MIN)/50 (R _MAX -R _MIN )/500

r ₁

(R _MAX -R _MIN )/50

r ₂

(R_MAX-R_MIN)/5 (R _MAX -R _MIN )/50

r ₂

(R _MAX -R _MIN )/5

CAL(k) is the calibration index.

R_NOM

1000；雖然應注意到此值及範圍係視所使用的特定模型具體價值函數而定及技藝人士能夠視所使用的模型而定組態R_NOM的適當範圍。在例示性的實施例中，r ₁ 可為從約1至約50的任何數，且r ₂ 可為從約10至約1000的任何數，較佳的是約500。雖然提供該調諧因子作為用於本文所述之實施例的特定數值之範圍，吾人注意到每個系統會採用不同調諧因子的配置以控制該控制器(和泵)在輸送胰島素時所需的回應。因此，吾人的意圖是不被束縛於任何特定的數值，且因此，如本文所述，該調諧因子可包括自一第一值或間隔至一第二值或間隔的任何值，其中該第一值或間隔將導致該泵對輸送胰島素具有非常快速(或高度激進)的回應，且該第二值或間隔將導致該泵對輸送胰島素具有非常慢(或保守)的回應。 Note that R _NOM can be a value from about zero to about 1000 (e.g. 0

R _NOM

1000; Although it should be noted that this value and range depend on the specific value function of the specific model used and the skilled person can configure the appropriate range of R _NOM depending on the model used. In an exemplary embodiment, r ₁ can be any number from about 1 to about 50, and r ₂ can be any number from about 10 to about 1000, preferably about 500. Although the tuning factor is provided as a range of specific values for the embodiments described herein, we have noticed that each system uses a different tuning factor configuration to control the response required by the controller (and pump) when delivering insulin . Therefore, our intention is not to be bound to any particular value, and therefore, as described herein, the tuning factor can include any value from a first value or interval to a second value or interval, where the first The value or interval will cause the pump to have a very fast (or highly aggressive) response to delivering insulin, and the second value or interval will cause the pump to have a very slow (or conservative) response to delivering insulin.

The current sample index is "k", where k _CAL is the most recent CGM calibrated sample index obtained using an appropriate reference glucose analyzer or glucose meter 114, and k _CAL2 is the next most recent one obtained using the reference glucose meter 114 The CGM calibration sample index. Therefore, 30 minutes (for example, 6 samples) after the calibration of the reference glucose meter 114, the calibration index module 32 or CAL(k) is zero. Therefore, CAL increases by one for each sample until 30 minutes after the next calibration. Until the next calibration, with each subsequent CGM measurement value of each time interval index k of the sensor 112, the correction of the tuning factor R of the module 34 at the instant "k" (ie R(k)) results in slightly more conservative The controller (that is, the higher R value of equation (4)). A 30-minute buffer is immediately followed after calibration. During this period, the control logic 10 will act relatively conservatively, because the CGM value immediately following calibration may "jump" significantly.

For additional details on how to adjust the tuning factor R for certain events (for example, signal loss or calibration), please refer to U.S. Patent Application No. SN13/708,032 (application date of December 7, 2012), which is in full here. Incorporated into this application as a reference.

In addition to adjusting the tuning factor for CGM signal reduction or calibration, we also invented a countermeasure. Whenever such a sensor is installed on the user's skin for the first time (or replaced by another), this countermeasure deals with the CGM sensor. The inaccuracy of the detector.

In summary, the system management object of diabetes in Figure 2A is provided. In this system, the following components are used: an intermittent glucose meter 29, a continuous glucose sensor 22, a pump 16, and a controller 10. The intermittent glucose meter 29 measures the subject's blood glucose at discrete and non-uniform time intervals (for example, every 4 hours) and provides the intermittent blood glucose level as a time interval index (k, where the time between k and k+1 The calibration index of each interval is 30. The continuous glucose monitor continuously measures the grape level of the subject at discrete approximately uniform time intervals (for example, approximately every 30 seconds or every minute) and provides the glucose level at each interval in the form of glucose measurement data, wherein at any interval Any omission of the glucose measurement is stored in omission index 32. The insulin infusion pump is controlled by the controller 10 to deliver insulin to the subject 20. As used herein, the term "insulin" is not limited to peptide hormones extracted from animals or genetically engineered insulin, biosynthetic recombinant human insulin or its analogs, but may include other drugs or organisms designed to mimic insulin. Preparations or other substances that have an effect similar to insulin in the human body. The controller 10 is programmed with an appropriate MPC program to control the pump and communicate with the glucose meter and glucose monitor. In this aspect, the controller determines the tuning factor 34 according to (a) the calibration index 30 derived from the intermittent blood glucose measurement value and (b) the omission index 32 of the MPC10 so that the controller predicts and controls the desired glucose concentration from the model according to (1) 12. (2) The glucose concentration 24 measured by the continuous glucose sensor 22 at each interval of the interval index (k) and (3) the tuning factor 34 to determine the insulin delivery rate at each time interval of the interval index (k) . However, I invented the system to change or replace the CGM sensor (the replacement of the sensor can be detected by the pump or confirmed by the user via a switch), as the specific CGM sensor is suitable for use. The inherent inaccuracy of the detector makes the system It is convenient to set the tuning factor R to a conservative value of, for example, 500 and continue for a predetermined time period, for example, 8 hours, 12 hours, 24 hours or longer or shorter.

2B and 2C, the telemetry output of the system is provided to verify the operation of the system. In Fig. 2B, the CGM output signal within 48 hours is displayed, and in Fig. 2C, the output of the tuning factor in the same time frame is juxtaposed at the bottom of Fig. 2B. In the event "a" of FIG. 2B, calibration is performed, which in a qualitative sense means that the output signal from the CGM sensor can be regarded as more accurate (because the intermittent glucose sensor or test strip is used for calibration). Therefore, in "b" of Figure 2C, MPC (located in the microcontroller) can increase the aggressive value of the tuning factor R (decrease its value to 50), and in the next two hours (from 12 hours to Within 14 hours), the aggressive value of the tuning factor R slowly ramps up to a larger value. In the event "c" of FIG. 2B, the microcontroller detects (or the user can confirm) that the CGM sensor has been replaced with a new CGM sensor from the flag or switch 36. The microcontroller (such as the programmer in one of the embodiments of our invention) immediately switches the tuning factor to a more conservative tuning factor, so as to make the insulin dosage more conservative. And for a predetermined time period (from 14 hours to about 40 hours), regardless of the signal drop from the new CGM sensor, this more conservative tuning factor is maintained for a predetermined time (for example, from 14 hours to about 40 hours). 40 hours). As used herein, the term "conservative tuning factor" refers to the controller's tuning factor R which is intended to ensure that the delivery rate of the insulin dose (relative to the benchmark in response to glucose changes for the planned application) is sufficiently low It also falls within the constraints of the specific application of the system to meet the security or regulation requirements for such applications. In a non-limiting example, the conservative tuning factor can be any value from about 500 to about 1000. However, when it is understood that this tuning factor can be any number used to affect the response or "gain" of a specific configuration of a system using a closed-loop controller (feedback, feedforward, or hybrid controller), as long as those familiar with this The technician believes that such tuning factors are safe or conservative.

Based on the disclosure provided herein, the applicant has also devised a method for controlling an infusion pump to control the infusion pump using a controller that controls the pump and receive data from at least one glucose sensor. With reference to Figure 3, an exemplary method will now be described. The method starts by measuring the glucose level of the subject from a continuous glucose sensor (22 or 29) in step 302, to provide at least one glucose measurement for each time interval of a series of discrete time interval indexes ("k") value. In step 304, the system queries whether the CGM sensor has been replaced within a predetermined time period X. The microcontroller can be used to check the sensor replacement log, the sensor replacement flag, or a combination of one or more signals indicating sensor replacement (for example, a manual actuation signal by the user). In the determination of the replacement of the sensor. If the query report is "true" or "yes", the system checks in step 306 whether the countdown timer for the predetermined time X is equal to zero (or whether the positive timer is equal to the predetermined time X). If it is false (ie, "No"), the system proceeds to step 308, where the tuning factor is set to a conservative value. In step 312, the microcontroller MPC now determined based on the cost J appropriate insulin dose zone I _D (from Equation 2-5), wherein the system cost J is affected by a tuning factor R. In step 314, the microcontroller activates the pump motor in the pump 102 to deliver the appropriate insulin dose calculated in step 312. In step 316, the microcontroller returns to the main program.

On the other hand, if the query report in step 304 is false (ie, "No"), the system maintains the tuning factor as the current setting before the sensor replacement event. Alternatively, the system can still calculate the appropriate tuning factor for the workload, as previously described in relation to equations 9-13.

Although the present invention has been illustrated by specific variations and illustrations, those skilled in the art can understand that the present invention is not limited to the variations or figures described. For example, the closed-circuit controller does not have to be an MPC controller, but if those skilled in the art can make appropriate modifications, the closed-circuit controller can be a PID controller, a PID controller with internal model control (IMC), and model calculations. Method Control (MAC), which was developed by Percival et al. in " Closed-Loop Control and Advisory Mode Evaluation of an Artificial Pancreatic β Cell: Use of Proportional-Integral-Derivative Equivalent Model-Based Controllers " (Journal of Diabetes Science and Technology , Vol.2, Issue 4, July 2008"); In addition, where the above methods and steps indicate specific events that occur in a certain order, those with ordinary knowledge in this art can understand and modify the order of certain steps And this kind of modification is based on the variation of the present invention. In addition, when feasible, they can be executed jointly in parallel programs, and some of these steps can be executed one after another as described above. Therefore, this patent intends to cover the equivalent variations that fall within the spirit of the disclosure or the scope of the patent application.

10: Control logic module; MPC logic module

12: desired glucose concentration; glucose concentration range

14: The first output; the control signal of the insulin pump

15: Predicted glucose value

16: insulin pump

18: The amount of insulin required

20: Object

22: Continuous glucose sensor

26: Difference

28: Update filter

29: Intermittent glucose meter

30: Calibration index module

32: data omission module

34: tuning factor

36: flag or switch

Claims (9)

A diabetes management system, wherein the diabetes management system is controlled by using a model predictive controller with a tuning factor (R), the tuning factor (R) is adjustable between a conservative tuning factor and a radical Among the tuning factors, the system includes using a microcontroller to control an infusion pump to control the infusion pump and receive data from at least one glucose sensor, and the method includes: each time interval in a series of discrete time intervals Within, the glucose level of the object is measured from the glucose sensor to provide at least one glucose measurement value; the determination of a change in the accuracy of the system includes: determining that the system is within a predetermined time interval A decrease in accuracy responds to the replacement of the glucose sensor by a new glucose sensor; an increase in the accuracy of the system is determined to respond to a calibration of the glucose sensor; settings for the micro-control The model predicts that the tuning factor (R) of the controller in the device is the conservative tuning factor in response to the decrease in the accuracy of the judgment in the system, or is set to the aggressive tuning factor in response to the judgment in the system The increase in accuracy; based on the model predictive controller using the tuning factor (R), the microcontroller calculates an amount of insulin for delivery to provide for one or more discrete time intervals A calculated amount of insulin to be delivered to the subject; and delivering the amount of insulin determined by the calculation step by activating the infusion pump by the microcontroller. Such as the system of the first item of the patent application, wherein the tuning factor (R) includes a value of about 500, and the predetermined time interval includes about 24 hours after replacing the glucose sensor with a new glucose sensor . Such as the system of the second item of the patent application, wherein the tuning factor (R) includes a value of about 10. Such as the system of the first item of the patent application, wherein the tuning factor (R) includes a value of about 1000. Such as the system of the first item of the patent application, wherein the conservative setting of the tuning factor includes a value of about 500. For example, the system in the scope of the patent application, wherein the tuning factor includes any value from a first value to a second or conservative value, the first value will cause the infusion pump to have a faster response to the delivery of the insulin , The second or conservative value will cause the infusion pump to have a slower response to delivering insulin, and the conservative setting includes the conservative value of the tuning factor. For example, the system of item 1 of the scope of patent application, wherein the model predictive controller has an error weighting factor (Q) related to the tuning factor (R), where:

Where R includes the tuning factor; Q includes an error weighting factor. Such as the system of the first item of the patent application, wherein the at least one glucose sensor includes a continuous glucose sensor and an intermittent glucose meter. A diabetes management system, wherein the diabetes management system is controlled by using a model predictive controller with a tuning factor (R), the tuning factor (R) is adjustable between a conservative tuning factor and a radical Among the tuning factors, the system includes using a microcontroller to control an infusion pump to control the infusion pump and receive data from at least one glucose sensor, and the method includes: each time interval in a series of discrete time intervals Inside, the glucose level of the object is measured from the glucose sensor to provide at least one glucose measurement value; to determine a change in the accuracy of the system, the determination includes: In a predetermined time interval, determine a decrease in the accuracy of the system in response to one or more of the glucose sensors being replaced by a new glucose sensor; determine an increase in the accuracy of the system To respond to a calibration of one or more glucose sensors, or to re-establish the glucose sensor to report glucose measurement; set the tuning factor (R) of the model predictive controller used in the microcontroller as the A conservative tuning factor in response to the decrease in the accuracy of the determination in the system, or the aggressive tuning factor in response to the increase in the accuracy of the determination in the system; based on the model using the tuning factor (R) The predictive controller calculates an amount of insulin to be delivered by the microcontroller, so as to provide a calculated amount of insulin to be delivered to the subject for one or more discrete time intervals; and through the micro-controller The controller activates the infusion pump to deliver the amount of insulin determined by the calculation step.

Images